Skip to main content
A social intelligence agent collects public social or media signals for research, monitoring, creator discovery, or content analysis. It should treat source records as evidence, model output as interpretation, and publishing or engagement as a separate authenticated action. This workflow is for source-bounded analysis, not unrestricted surveillance or automatic posting.

​
Choose the capability

Use an Agent Skill when the monitoring or analysis method is reusable. Use direct APIs when the application must control queries, pagination, deduplication, scoring, storage, or source-specific logic.

​
1. Define the research question

Specify the topic, entities, time window, platforms, languages, and output needed. Avoid collecting broad social data without a clear purpose. Examples include:
  • Track how a product launch is discussed across selected public sources.
  • Identify recurring questions in a defined community.
  • Compare creator coverage of a topic during a stated time window.
  • Find source material for a content brief.

​
2. Collect source-bounded records

Use the smallest relevant APIs and preserve:
  • Source platform and canonical URL.
  • Author or channel identifier when needed for the task.
  • Publication and retrieval times.
  • Query and filter parameters.
  • Raw metrics returned by the source.
Do not treat missing results as proof that no discussion exists.

​
3. Normalize without erasing provenance

The application may normalize timestamps, identifiers, engagement fields, and text for analysis. Keep the original source reference so each normalized record remains auditable. Separate:
  • Source-provided fields.
  • Application-derived metrics.
  • Model-generated labels or summaries.
  • Missing, inaccessible, or uncertain data.

​
4. Analyze patterns cautiously

A model can cluster themes, compare narratives, summarize recurring questions, or identify candidate signals. It should not invent audience demographics, intent, sentiment, identity, or causal relationships that the sources do not establish. For important conclusions, compare multiple records and, when possible, more than one source type.

​
5. Produce an evidence-linked brief

A useful result should include:
  • Scope and time window.
  • Sources and collection method.
  • Observed themes or changes.
  • Representative links or records.
  • Model interpretation labeled separately.
  • Gaps, sampling limits, and uncertainty.

​
6. Gate external actions

Reading public social data does not authorize posting, liking, following, replying, messaging, or changing an account. Before an external action:
  1. Verify the connected account and required permission.
  2. Show the exact content, target, and intended effect.
  3. Obtain confirmation where policy requires it.
  4. Execute the narrowest action once.
  5. Verify the resulting external state.

​
Privacy and safety boundaries

  • Collect only data needed for the defined purpose.
  • Respect platform terms, applicable privacy requirements, and audience expectations.
  • Avoid inferring sensitive traits from social activity.
  • Do not expose credentials, private account data, or unrelated personal information.
  • Treat automated engagement as a write operation with reputational and compliance risk.

​
Minimal documentation set

​
When AIsa may not be necessary

A direct platform integration may be simpler when the application uses one source, already has stable authentication and compliance controls, and does not need models, reusable Skills, or other AIsa resources.